A Key Word History of Marketing Science Carl F. Mela Jason M.T. Roos Yiting Deng 1 November 5, 2012 1 Carl F. Mela is the T. Austin Finch Foundation Professor of Business Administration at the Fuqua School of Business, Duke University (email: [email protected], phone 919-660-7767). Jason M.T. Roos is an Assistant Professor at the Rotterdam School of Management, Erasmus University (email: [email protected], phone 206-317-1713). Yiting Deng is Ph.D. candidate at the Fuqua School of Business, Duke University (email: [email protected], phone 510-735-6255). The authors would like to thank seminar participants at the 2012 ISMS and AMA Doctoral Consortiums, the University of Houston, and the University of Frankfurt, as well as Bart Bronnenberg, Preyas Desai, Rick Staelin, and an anonymous reviewer for their feedback.
Transcript
A Key Word History of Marketing Science
Carl F. Mela Jason M.T. Roos Yiting Deng1
November 5, 2012
1Carl F. Mela is the T. Austin Finch Foundation Professor of Business Administration at theFuqua School of Business, Duke University (email: [email protected], phone 919-660-7767). JasonM.T. Roos is an Assistant Professor at the Rotterdam School of Management, Erasmus University(email: [email protected], phone 206-317-1713). Yiting Deng is Ph.D. candidate at the Fuqua Schoolof Business, Duke University (email: [email protected], phone 510-735-6255). The authorswould like to thank seminar participants at the 2012 ISMS and AMA Doctoral Consortiums, theUniversity of Houston, and the University of Frankfurt, as well as Bart Bronnenberg, Preyas Desai,Rick Staelin, and an anonymous reviewer for their feedback.
Abstract: A Key Word History of Marketing Science
This paper considers the history of key words used in Marketing Science to develop in-
sights regarding the evolution of marketing science. Several findings emerge. First, pricing
and game theory are the most ubiquitous words. More generally, the 3C’s and 4P’s predom-
inate, suggesting key words and common practical frameworks align. Various trends exist.
Some words, like pricing, remain popular over time. Others, like game theory and hierar-
chical Bayes have become more popular. Finally, some words are superseded by others, like
diffusion and social networking. Second, the overall rate of new keyword introductions has
increased but the likelihood they will stick around has decreased. This suggests a matura-
tion of the discipline or a long tail effect. Third, a correspondence analysis indicates three
distinct eras of marketing modeling, comporting roughly with each of the past three decades.
These eras are driven by the emergence of new data and business problems, suggesting a
fluid field responsive to practical problems. Fourth, we consider author publication survival
rates which increase up to six papers and then decline - possibly due to changes in ability or
motivation. Fifth, survival rates vary with the recency and nature of words. We conclude
by discussing the implications for additional journal space and the utility of standardized
classification codes.Keywords: Key Words; History; Marketing; Data Visualization.
1
1 Introduction
We characterize the history of marketing science using the key word choice of authors in
Marketing Science. Authors’ choices of key words not only afford a sense of what topics
were once, have always been, or are becoming central to marketing thought; they also reflect
factors that drive innovation in the field. That is, the collective efforts of the field, as
expressed by the key word choices of its authors, provide a sense of the field’s interests
and the potential for subsequent innovations. This research also complements a strategic
review of the journal, appearing in this issue of Marketing Science, regarding stakeholders’
perspectives about the journal’s goals and focus Chintagunta et al. (2013). Our findings
about the state of the field, and how it has evolved over time, can be viewed as a potential
input to such a strategic assessment. For example, we consider i) the interplay between
methods and topics and between research and practice and ii) the rate at which seminal
topics have been introduced over time - issues central in that strategic review.
Other research exists with similar aims. These studies can be characterized as either
survey based (using the authors’ review of the literature) or citation based (using citation
analyses). For example, Leeflang et al. (2000) characterize five eras of marketing science,
the latter three of which are spanned by the publication of Marketing Science (see Lilien
(1994), for another summary). Our analysis also suggests three eras of marketing science that
loosely comport with their characterizations. Related, Buzzell (1968) and Lilien et al. (1992)
developed definitive texts outlining conceptual foundations for the field of marketing science.
In contrast, our work is more empirically grounded. Along these lines, there also exists a
broad literature on citation analyses in marketing (e.g., Baumgartner and Pieters, 2003;
Tellis et al., 1999). By assessing author cross-citations, these papers assess the similarity
between journals. More recently, Moussa and Touzani (2010) use Google scholar citations
to assess the relative impact of journals. In contrast, we focus our attention on the impact
of topics and assess how this varies over time.
2
We are aware of one marketing paper that also analyzes key words. Kevork and Vre-
chopoulos (2009) consider key words used in customer relationship management research,
using expert judgment to classify these key words into topics. Our approach extends this
work by i) considering the broader domain of marketing models, ii) applying multivariate
analyses for classification, and iii) considering dynamics in word usage.
Several findings emerge. First, the modal use of a key word is just once, perhaps reflecting
a long-tail for key words. Second, regarding the words that are most prevalent, there is a
correspondence between the common marketing vernacular of the “3C’s” and “4P’s” and
the most predominant words used. Third, key words tend to cluster into several themes,
including promotions, channels, and new products. Of note, “game theory” is atop this
hierarchical clustering, being the word most likely to connect these various threads. Fourth,
patterns exist in the prevalence of key words over time, with some words in ascendance
(like “game theory”), some in maturity (like “promotions”) and some being substituted with
newer concepts (like “diffusion” and “social networks”). Fifth, while the rate at which new key
words appear is accelerating, the rate at which they become enduring is decelerating, possibly
suggesting a long-tail phenomenon wherein the additional inventory of journal space leads
to more variety in key words. Finally, we conclude with an analysis of author productivity
rates, in which we observe that about 3/5 of all authors who publish in Marketing Science
do so just once. Moreover, the probability of continuing to publish peaks around the fifth
paper, indicating longevity per se does not explain survival rates in publication. Of note,
these survival rates are related to the choice of words (methods or topics), their novelty, and
authors’ tendencies to use different words on new papers.
The remainder of this note proceeds as follows. First, we detail the key word and au-
thorship data used in our analysis. Next, we outline our approach for analyzing these data.
We conclude by summarizing key insights and their potential implications.
3
2 Data
Data used in the analysis comprise all articles published in Marketing Science from its
inception in 1982 through 2011 - 30 volumes in total. For each article, the data include
the authors, volume, issue and date of publication, title, and the sets of key words chosen
by the article’s author(s). There are no restrictions used on key words that authors can
select in Marketing Science nor are there any ex ante classifications of key words from which
to choose (c.f. the JEL Classification Codes used by many economics journals, which we
discuss below). All totaled, there are 4749 key word choices, of which 2415 are unique,
describing 1085 articles by 1051 authors (or co-authors). Given our interest lies with the
organic evolution of the field, rather than policy, all 35 editorials appearing in the data are
excluded, leaving 1050 articles, 1050 (co-)authors, and 4654 key words (of which 2357 are
unique).
One obstacle to analyzing these data arises from conceptual overlap in key words with
slightly different spellings, e.g., “auction” and “auctions;” “word of mouth” and “word-of-
mouth;” and “variety seeking,” “variety-seeking,” and “variety-seeking behavior.” Accord-
ingly, highly similar terms are merged. Likewise, we combine synonyms such as “store brands”
and “private labels,” and “movies” and “motion pictures” using our collective judgment for
these decisions. These substitutions are listed in an online technical appendix. After com-
bining words, the data reduce further to 1975 unique key words.
Remarkably, most of the key words appearing in Marketing Science have been used
just once, and 97% of papers include at least one key word never before used. The mean
and median number of key word appearances over more than three decades are 2.3 and 1
respectively. This suggests that most key words never “stick.”
4
3 Analysis and Historical Insights
3.1 Top Words
Table 1 reports the frequency count of the top 25 words used in Marketing Science over
the past three decades. “Pricing” appears most often. Of note, elements from the so-called
“4P’s” - “pricing;” “advertising” and “promotion;” “channel” and “retailing” (place); and “new
products” - occupy six of the top eleven slots, demonstrating the centrality of this framework
to the field and providing evidence for a strong alignment between marketing science and
marketing practice. Likewise, aspects of the so-called 3C’s are also heavily represented in the
leading words, with “choice models” (customer) and “competitive strategy” in the top six.
The impact of methodological innovation is evidenced in Marketing Science; over a quarter
Table 1: Most frequently used key words in Marketing Science, 1982 - 2011. Among thewords used more than 40 times, the 4P’s are marked with a closed triangle (a) and the 3C’swith an open triangle (1).
The appearances of individual key words are not independent. To understand how these
words relate to each other, we perform a hierarchical cluster analysis using the Jaccard (1901)
distance metric, defined as the ratio i) of the number of times both keywords appear in the
5
same article to ii) the total number of articles featuring either keyword, #(A∩B)/#(A∪B).
This metric classifies words as more similar if they appear alongside many of the same
key words. The dendrogram in Figure 1 depicts the resulting clusters for the top 20 key
words. Distances between leafs in the dendrogram, i.e., the sum of the lengths of the stems
separating any two leafs, depict the (dis)similarity of words. For example, “Bayesian analysis”
and “hierarchical Bayes” are closely related because they tend to appear with the same sets
of other words.
Customer satisfaction
Price discrimination
Advertising
Bayesian analysis
Hierarchical Bayes
Conjoint analysis
Forecasting
New products
Diffusion
Brand choice
Promotion
Buyer behavior
Econometric models
Choice models
Channel
Competition
Competitive strategy
Pricing
Retailing
Game theory
Figure 1: Dendrogram depicting a cluster analysis of the 20 most prevalent keywords. Groupsof similar key words are contained in the same boxes.
Four key groupings are suggested by Figure 1. The words in the lowest box comprise the
most specialized cluster, which represents words not as frequently used. In the lower level of
this cluster, we observe “Bayesian methods,” “price discrimination,” “customer satisfaction,”
and “advertising,” which may reflect customer-level targeting. At this cluster’s highest level,
“diffusion,” “new products,” “forecasting,” and “conjoint analysis,” are focused on new product
demand, and the upper level’s link to the lower level of the cluster may lie in the concept
6
of demand assessment. The next box up includes “econometric models,” “brand choice,”
and “choice models,” possibly reflecting the scanner data era of research. The second box
from the top includes “channel,” “competition,” and “pricing,” and appears to reflect research
pertaining to channel strategy. Finally, of special interest is the key word “game theory” (at
the top of Figure 1), which is not closer to any one topic cluster than another, but rather
spans them all. Applied initially in the context of analytic models, “game theory” can be
construed as a unifying lens through which researchers look in order to study a plethora
of issues in marketing science. This trend has been extended to empirical research in the
context of structural models, which have also adopted game theoretic techniques. The theme
of how keyword use evolves over time is a point we consider next.
3.2 Key Word Lifecycles
Figure 2 reports word clouds for all keywords occurring over three 10 year periods of our
data, aligning roughly with the 1980’s, 1990’s and 2000’s. In the word cloud, the font size of
the word is proportional to its frequency of use over that duration. For example, a dominant
word in each decade is “pricing,” consistent with Table’s 1 result that “pricing” is the most
heavily used keyword.
7
1982-1991 1992-2001
2002-2011
Figure 2: Word cloud by decade. Words that were used more frequently in a particulardecade are shown in larger type.
Figure 2 suggests that the use of key words is not static; some have always been popular
(e.g., “advertising,” “pricing,” “new products,” and “promotion”), some are in growth (e.g.,
“game theory” and “hierarchical Bayes”) and some are in substitution (“diffusion” was popular
in the 1980’s, but “social networks” and “Internet marketing” are increasingly popular in the
2000’s). The figure also suggests the concentration of the leading key words by decade; there
are a smaller percentage of very popular words in Marketing Science’s first decade, perhaps
indicating that the field is in the growth phase of its product life cycle (Levitt (1965)) and
becoming more fragmented over time (in terms of words used).
Figure 3 plots the share of selected key words over time in order to illustrate some of the
more prominent trends in the data: growth, maturity and substitution. In panel (b) and
(d) on the right, one observes the rapid ascendance of game theory and Bayesian methods
as prominent paradigms in marketing. In panel (a) of Figure 3, we see that “pricing” and
“promotion” were especially popular in the 1990’s, likely a consequence of the ubiquity of
8
scanner data research at the time and suggestive of maturity. The substitution of “social
networks” and “Internet marketing” for “diffusion,” which we noted earlier, is also evident in
panel (c).
0%
1%
2%
3%
4%
5%
6%(a)
Sha
re
1982−1984
1985−1987
1988−1990
1991−1993
1994−1996
1997−1999
2000−2002
2003−2005
2006−2008
2009−2011
AdvertisingPricingPromotion
0%
1%
2%
3%
4%
(b)
Sha
re
1982−1984
1985−1987
1988−1990
1991−1993
1994−1996
1997−1999
2000−2002
2003−2005
2006−2008
2009−2011
ChannelCompetitionGame theory
0%
0.5%
1%
1.5%
2%
2.5%(d)
Sha
re
1982−1984
1985−1987
1988−1990
1991−1993
1994−1996
1997−1999
2000−2002
2003−2005
2006−2008
2009−2011
Bayesian analysisHierarchical Bayes
0%
1%
2%
3%
4%
5%(c)
Sha
re
1982−1984
1985−1987
1988−1990
1991−1993
1994−1996
1997−1999
2000−2002
2003−2005
2006−2008
2009−2011
DiffusionInternet marketingSocial networks
Figure 3: Key word popularity over time. The four panels illustrate patterns of growth andsubstitution.
3.3 Innovation and What Sticks
Figure 4 depicts a scatter plot of key words by time. The vertical axis lists key words in
the order of their first appearance in the data (earlier appearing words are at the bottom).
The words used are suppressed to conserve space. The horizontal axis represents the years
in which these keywords first appeared. Each circle on the plot represents the appearance
of a word in a year, and the size of the circle is proportional to a key word’s share in that
same year. Editor names are listed across the top of the figure for reference.
9
1985 1990 1995 2000 2005 2010
Donald G. MorrisonDonald G. MorrisonSubrata K. SenSubrata K. Sen John R. HauserJohn R. Hauser Richard StaelinRichard Staelin Brian T. RatchfordBrian T. Ratchford Steven M. ShuganSteven M. Shugan Eric T. BradlowEric T. Bradlow
Preyas DesaiPreyas Desai
Key
wor
ds
Years
Figure 4: Emergence and popularity of new key words by year. Rows represent individualkey words. Circle size is proportional to key word share in each year. Colored bands groupkey words together by the year they first appeared.
One insight emerging from the figure is that some words have been quite enduring. Many
words appearing in 1992, for example, continue to appear today. Key words from other years
(e.g., 1989 or 2004) have been less enduring. Denser horizontal “bands” reflect eras during
which new, enduring key words were established (it would be of interest to assess what
aspects of these eras led to their more enduring word use). Another insight is that the pace
of key word introduction, as evidenced by the slope of the key word frontier, is accelerating.
Much of this acceleration can be ascribed to an increase in journal space in 2004. The slope
of the frontier is roughly linear both before and after this increase (though may have leveled
some in the mid 1990’s). This acceleration might indicate the field has entered a growth
phase, suggesting it has never been easier to innovate in terms of key words (and topics
to the extent these equivalate with words). Another potential explanation for the patterns
10
observed is that there is a long-tail phenomenon with key word choice, wherein a core set
of papers continues to have impact, but most of the growth in the field arises from papers
that are marginally less impactful (Anderson, 2006). If so, one might expect - as we see in
the data - that the increase in the number of new words would not be accompanied by an
Figure 5: Emergence, persistence, and decline of highly popular key words, grouped in three-year periods. Words are ordered by their popularity in the year they first broke into the top10. Circle size represents popularity within each year.
A similar scatter plot is produced in Figure 5, but now focuses solely on the top ten most
popular key words in each of ten three year periods (we use three year increments in lieu of
one year increments to facilitate interpretation, and include the top 11 words when there is
a tie). It is important to note that the first appearance of a word on this plot marks the
time at which a word first became popular enough to enter the top 10 in market share, and
11
not the first time the word appeared in the literature. For example, even though Coughlan
(1985) is the first paper published in Marketing Science to use the key word “game theory,”
the topic did not become popular until the mid 1990’s. That is, there can be a response
latency between a word’s first appearance and when it becomes popular. Of note, Figure 5
reinforces the notion that some words become enduring (e.g., “game theory,” “channel,” and
“competitive strategy”), whereas some words are more ephemeral (e.g., “variety seeking” and
“market segmentation”).
Figures 4 and 5 present an informative contrast: Figure 4, which includes all key words,
is convex, whereas Figure 5, which includes only the most popular key words, is concave.
Apparently, while it has become easier to innovate, it has also become more difficult to
become (and remain) popular. This is consistent with the product life cycle notion that
the field may be transitioning from growth to maturity (Klepper, 1996). This result is also
suggestive of a long tail aspect to the increase in journal space. Though it is impossible to
answer this question with these data (for example, words might not equate to topics and
there is a latency between their first appearance and when they become popular), the issue
merits additional consideration.
Regarding the concern that words might not align well with topics, one way to assess
this is the preceding cluster analysis. Another approach, which further considers dynamics,
is correspondence analysis (Hoffman and Franke, 1986), which we discuss next.
3.4 Eras of Marketing
In our application of correspondence analysis, years and keywords are colocated based on
the frequency with which the words appeared, as depicted in Figure 6. One can think of
a year being placed in the center of the words that were used that year, and a key word
being placed in the center of the years in which it appeared. In order to keep the figure
interpretable, we focus on words occurring at least 10 times. Results are reported in Figure
6.
12
Figure 6: Correspondence analysis showing key words and years. Key words are placedclosest to the years in which they were the most popular.
As one might expect, terms that are used in all years, like “pricing,” appear closer to the
center of the plot. These terms are less informative about the evolution of the field than
those located closer to the periphery, which are more representative of words used in spe-
cific eras. Three such eras emerge, roughly corresponding to the decades of analysis, 1980’s
(located on the right), 1990’s (top), and 2000’s (bottom). In the 1980’s, “consumer choice”
and “logit” were especially popular, reflecting the promotional tools developed in that era
(e.g., Guadagni and Little, 1983). In the 1990’s, scanner data became widely available and
accordingly emphasis shifted toward work developed around “brand choice” and “economet-
13
ric models,” as further evidenced in 3. In the 2000’s, we see the emergence of “Bayesian
methods,” “Internet,” “CRM/targeted marketing,” and more recently, “social networks.” The
evolution from aggregate models of brand choice to individual level models of social engage-
ment suggests that the field is reacting strongly to the informational environment in which
it operates. As new industries and data (e.g., log files of customer interactions from fields
such as the Internet, telecom and banking) become increasingly available, the field responds
in terms of tools and topics. Arguably, 6 stands in contrast to the perspective of some that
the field is becoming less relevant (Lehmann et al. (2011)).
3.5 Author Publications
In this section we consider author productivity and mortality at Marketing Science, and the
potentially moderating role played by key words.
3.5.1 Author Productivity
The publication data are indicative of authors’ research productivity, allowing one to assess
norms within the field. Figure 7 plots the distribution of publications per author in panel
(a) on the left and rate of publications in panel (b) on the right. The latter is defined
as the total number of publications divided by the number of years between the first and
last publication observed. These figures are conditioned on having published in Marketing
Science, a challenging task to begin with. Panel (a) indicates that authors who publish more
than once are in the top 40% of the field in their productivity, reflecting the challenging
nature of publishing in an elite journal. Panel (b) suggests that those who publish two
or more papers do so at a rate of about two papers per year (with the caveat that these
observations are skewed towards authors with fewer than five papers). Overall, the message
is that it is difficult to publish, but that once this is done, publication rates become somewhat
more regular.
14
0 5 10 15 20 25 0%
10%
20%
30%
40%
50%
60%S
hare
(a) Number of Publications0 1 2 3 4
0%
5%
10%
15%
Sha
re
(b) Number of Publications Per Year
Figure 7: Distribution of research productivity. The histogram in panel (a) shows theproportion of authors publishing a certain number times in Marketing Science. Panel (b)shows the proportion of authors achieving a given publication rate.
Exploring this idea in greater detail, Figure 8 depicts the author survival rate, defined
as Pr[∑K
k+1 nk > 0|nk
]where nk = 1 if an author publishes k papers and 0 otherwise. One
insight is that it becomes easier to publish up until 6 papers, at which point the likelihood
of publishing an additional paper reaches 80%. This increase might reflect increasing skill
or possibly selection bias. The likelihood of publishing then declines slowly, potentially
owing to either ability (as skills become obsolete) or motivation (as tenured researchers
pursue alternative endeavors). One noteworthy aspect of these figures is that it is difficult to
publish, and remains so over the course of a career; there is no evidence of an “inside club.”
15
1 2 3 4 5 6 7 8 9 10 11 12 13 150
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
Current Number of Papers
Con
ditio
nal P
roba
bilit
y of
Hav
ing
Mor
e P
aper
s
Figure 8: Author survival rates. Bars indicate the probability of publishing an additionalpaper, given a certain number have already been published.
3.5.2 State Dependence and Author Survival
Systematic differences exist among survival rates for authors depending on the types of key
words used (e.g., methods vs. topics, growth vs. decline), and authors’ tendencies to repeat
the same key words across many papers. Accordingly, we integrate our author survival and
key word analyses. To achieve this aim, we first compute, for each author-paper-key word
observation, the following set of dichotomous variables: i) whether the author used the same
key word on their next paper (repeat), ii) whether they did not use the same key word
on their next paper (switch), or iii) whether the author who used that key word published
another paper (survival). Then, for each key word, we computed an empirical probability
(across author-papers) of repeating the key word, switching to a new key word, or never
publishing again. Note that these three state variables sum to one, and that the probability
of survival after using a key word is the sum of that word’s repeat and switch probabilities.
16
0 5 10 15 20 25 3030
35
40
45
50
55
60
65
Pricing
Game theoryAdvertising
Choice models
Channel
Competitive strategy
Econometric models
PromotionNew productsBrand choice
Retailing
Competition
Bayesian analysis
Diffusion
Hierarchical Bayes
Buyer behavior
Forecasting
Conjoint analysis
Customer satisfaction
Price discrimination
Market structure
Marketing mix
Marketing strategy
Structural models
Internet marketing
Motion picture
Percent Repeating
Per
cent
Sw
itchi
ng
Top 25 Words
MethodTopic
Figure 9: State Dependence in Key Word Usage. Circles indicate the relative usage of keywords. Larger circles denote greater use. Solid circles indicate methods key words, opencircles indicate topics key words. The dashed lines are iso-curves indicating regions withconstant survival rates. The lowest iso-curve corresponds to a 35% survival rate, the highestto 60%. Observations in the lower left (upper right) indicate lowest (highest) survival.
We compare the resulting repeat, switch, and survival probabilities for the top 25 key
words in Figure 9. Several insights emerge. First, authors using methods words tend to have
higher repeat likelihoods and lower switch likelihoods. This finding might be ascribed to the
use of these tools as over-arching paradigms to study a variegated mixture of substantive
topics that change from paper to paper. Greater inertia in methods may be indicative
of higher barriers to entry for methods relative to topics (e.g., learning costs). Second,
more common words, as evidenced by circle size, tend to be repeated more often - as one
might expect owing to their increased prevalence. Third, words that are in their ascendancy
(“Game Theory, “Bayesian Analysis”, and “Internet Marketing”) relative to those that are
not (“Forecasting” and “Market Structure”) are also words that are high in repeat and low in
17
switching.1 Such a pattern might arise if authors were to focus on “hot” issues by publishing
runs of paper on these topics. In contrast, older words are more likely to have switches
because it is harder to publish in declining areas.2
Finally, it is possible to link research topics to survival. If all topics had the same survival
rate, but differed only on the trade-off between repeat and switch, then all 25 words would
perfectly align along a negative 45 degree slope, as indicated by the red dashed lines. In
fact, a least square error fit of a line to the observations yields an intercept of 0.58 and a
slope of -0.94. This result suggests that the slope of the observations in Figure 9 is indeed
very close to -1, and that key words lie on an iso-survival curve of 60%. However, there
is substantial dispersion from this line, as indicated by an R2 of 0.29. This dispersion is
greatest for the observations on the left hand side of the Figure 9 where repeat rates are
low. As discussed earlier, words in the left tend to describe older topics. Two strategies are
thus evidenced. First, above the 60% survival iso-contour, words such as “market structure”
and “price discrimination” indicate higher survival rates. This pattern is consistent with
survival by switching to new topics. Second, for those older words below the 60% iso-
survival contour (such as “marketing strategy”) lower switching corresponded with lower
survival rates. Interestingly, there is less dispersion of key words around the 60% survival
iso-survival curve in the right hand side of the figure. These words evidence higher repeat
and lower switching. Given that many of these words are method-oriented, one possibility
is that authors use a method as a base, switching topics as time passes.
4 Conclusion
This paper identifies trends in marketing science through the lens of key words used in
Marketing Science. The analysis first and foremost helps to document the history of the1Note the similarity of the north-west to south-east orientation of word order in Figure 9 to the north-east
to south-west orientation of word order in Figure 6). This result is consistent with the interpretation of wordsin growth and decline over the decades.
2This finding pertaining to new words could also relate to censoring in that new words have yet to entera decline that would incent authors to switch. To explore this possibility we created the same plot whileomitting observations from the past five years. The results appear qualitatively similar.
18
journal, and in that sense may be a useful input in constructing a situation analysis for
assessing the state of the journal. Moreover, the history of Marketing Science is potentially
of interest in its own right to the authors who helped develop it.
The analysis generates a number of insights. First, the foundational aspects of marketing,
as construed via the 3C’s and 4P’s, comport with the most frequent words used, suggesting a
tight link between commonly used frameworks for defining the field and the choice of research
questions considered. Second, the field is dynamic, with patterns of growth (e.g., “game
theory”) and substitution (e.g., “diffusion” and “social networks”). This is an encouraging
sign that the field is actively evolving. In the case of “game theory,” we further find that it
does not cluster with other words; in this sense it is more of a unifying theme/paradigm that
enables researchers to find common ground across topics and methods, like the “one ring
to rule them all” (Tolkien, 1954). Third, there is a direction to this evolution that mirrors
the evolution of data and issues faced in practice, indicating that the field does not appear
to be veering from its applied roots. More data would be useful to better understand how
trends in practice drive trends in research - and whether applied topics lead research, or
research leads applied topics. Fourth, inspection of publication rates reveals that while it is
challenging to publish, it becomes somewhat easier up to a point - and that the choice of
key words seems related to the likelihood of survival. Last, survival rates may be affected
by the types of words chosen and the authors likelihood of repeating them.
In addition, there are a number of relevant policy questions that emerge from this analysis.
Owing in part to the addition of more journal space in 2004, it has become easier to innovate
with key words. This is a good thing, in the sense that it is potentially enabling new ideas
and might reflect a field in growth. Yet this growth in innovation has been matched by a
deceleration in the rate with which new words stick. This might indicate a field in maturation,
or perhaps reflect that more space is coincident with a longer tail in which the marginal papers
published are not likely to be as influential as the infra-marginal papers. The latter might
suggest that there are diminishing returns to adding additional journal pages. The decrease
19
in rates of new popular words should also be recognized by tenure committees looking for
“home run” papers. The data suggest that the likelihood of publishing such papers over time
might be declining.
A second policy question pertains to the use of classification codes such as the Journal of
Economic Literature (JEL) codes used by the American Economic Association. In Marketing
Science, it has been possible to use any key words, and in many years, there has been no
code list. There are several advantages to using such a list. First, standardizing key words
makes it easier to link words to topics. In our analysis, because words are often similar,
words provide a noisy measure of topics, making it difficult to track the evolution of the
field in terms of topics and the relative impact of ideas. Second, key words make it easier
for authors to find similar works, and therefore encourage relevant early papers to diffuse
into subsequent papers. For example, we aggregated several words into “motion picture,”
including “film production,” “films,” “motion picture distribution and exhibition,” “motion
picture industry,” “movie industry,” “movie theaters,” and “movies.” Papers using “films”
might be harder to find by those using “movies” or “motion picture” even though these are
synonyms. Third, key words enable editors to match reviewers and papers more easily, as
one can consider the overlap between the papers and the authors that publish those papers.
On the negative side, if the field is moving quickly, it might be somewhat constraining to
limit organic growth of words as early indicators of new topics - i.e., it would be difficult for
a committee tasked with key word classification to foresee all new topics as they first emerge.
Hence a hybrid system might be of value (e.g., requiring most key words be chosen from a list,
while allowing a limited number of new words). It seems sensible to explore these tradeoffs
in greater depth and assess the suitability of classification codes for Marketing Science.
In sum, it is our hope that this analysis of the history of Marketing Science key words
provides a portrait of the field that is of historical value and has implications going forward
- and that this approach might be more generally applied in the context of other journals to
obtain a broader understanding of this and other fields.
20
ReferencesAnderson, Chris. 2006. The Long Tail: Why the Future is Selling Less of More. Hyperion.
Baumgartner, Hans, Rik Pieters. 2003. The structural influence of marketing journals: Acitation analysis of the discipline and its subareas over time. Journal of Marketing 67(2)pp. 123–139.
Buzzell, Robert D. 1968. Mathematical Models and Marketing Management . Harvard Uni-versity Press.
Chintagunta, Pradeep, Dominique Hanssens, John Hauser, Jagmohan Singh Raju, KannanSrinivasan, Richard Staelin. 2013. Marketing science: A strategic review. MarketingScience forthcoming. 1–.
Coughlan, Anne T. 1985. Competition and cooperation in marketing channel choice: Theoryand application. Marketing Science 4(2) pp. 110–129.
Guadagni, Peter M., John D. C. Little. 1983. A logit model of brand choice calibrated onscanner data. Marketing Science 2(3) pp. 203–238.
Hoffman, Donna L., George R. Franke. 1986. Correspondence analysis: Graphical represen-tation of categorical data in marketing research. Journal of Marketing Research 23(3) pp.213–227.
Jaccard, Paul. 1901. Distribution de la flore alpine dans le bassin des drouces et dans quelquesregions voisines. Bulletin de la Société Vaudoise des Sciences Naturelles 37(140) 241–272.
Kevork, Eleni K., Adam P. Vrechopoulos. 2009. Crm literature: Conceptual and functionalinsights by keyword analysis. Marketing Intelligence and Planning 1(1) 48–55.
Klepper, Steven. 1996. Entry, exit, growth, and innovation over the product life cycle. TheAmerican Economic Review 86(3) pp. 562–583.
Leeflang, Peter S.H., Dick R. Wittink, Michel Wedel, Philippe A. Naert. 2000. Buildingmodels for marketing decisions . Boston: Kluwer Academic Publishers.
Lehmann, Donald R, Leigh McAlister, Richard Staelin. 2011. Sophistication in research inmarketing. Journal of Marketing 75(4) 155 – 165.
Levitt, Theodore. 1965. Exploit the product life cycle. Harvard Business Review 43(6)81–94.
Lilien, Gary L. 1994. Marketing models: Past, present and future. Research Traditions inMarketing , chap. 1. Kluwer, 1–20.
Lilien, Gary L., Philip Kotler, K. Sridhar Moorthy. 1992. Marketing Models . Prentice Hall,Englewood Cliffs, NJ.
Moussa, Salim, Mourad Touzani. 2010. Ranking marketing journals using the google scholar-based hg-index. Journal of Informetrics 4(1) 107 – 117.
Tellis, Gerard J., Rajesh K. Chandy, David S. Ackerman. 1999. In search of diversity: Therecord of major marketing journals. Journal of Marketing Research 36(1) pp. 120–131.
21
Tolkien, John R.R. 1954. The Lord of the Rings . George Allen & Unwin.
